Transcript Wood

Assessing the Quality of SPPIs
M G Šova, J Wood & I Richardson
Office for National Statistics, UK
Overview
• What are SPPIs?
• Quality Assessment of ONS SPPIs
– Products
– Business versus retail
– Sampling
– Weighting
• Non-standard estimation methods
• Other aspects
• Plans to improve and monitor quality
What are Services Producer Price Indices?
Price movements in business-to-business services
In UK:
– Quarterly Laspeyres price index
– Rebased every 5 years
– Uses:
• Estimating corporate services inflation
• Deflators for Index of Services
– Users:
• National Accounts
• Bank of England
• Eurostat
Quality Assessment of ONS SPPIs
• In 2006 by Dr Markus Šova
– Replaced planned three-year programme
• 29 Product Classes
– 198 Product Categories
• 134 recommendations
Quality Assessment of ONS SPPIs
• Product descriptions often vague
• Dubious sample allocations between categories
• Weights
– Some excessively large (potentially disclosive)
– Some excessively small
– Derivation and rationale not always clear
– Weighting for purposively selected respondents problematic
• Coverage incomplete
• Definition of ‘Europe’ ill-defined
• Some uses of external indices problematic
• Processing risks (manual transcription, spreadsheets)
Products - Classification
• CPA 2008 :
124 pages
– Agriculture & construction: 12 pages
– Production:
72 pages
– Wholesale & retail:
9 pages
– Other services:
31 pages
• In-house service product classification
Product Definitions
• Need consistent product definitions over time
• Not always possible (e.g. Market Research)
• Not always possible to define the product
(e.g. canteen services)
• Possible Solutions:
– standard services
– quality adjustment
– hedonic modelling
– model contracts
– hourly rates
– unit value prices
Business versus Retail
• Intermediate consumption v. final consumption
(Supply-Use Tables)
• Differentiation not always possible (e.g. rail fares)
• Possible solution:
– Product differentiation and differential weighting
(e.g. by destination, travel time, ticket class)
– Data not always available
Sampling
• ProdCom (sample size: 22,000)
• Ad hoc SPPI turnover surveys in 2000 and 2005
– Sample size: 5,000
(SPPI sample size: 4,500)
– Limited scope and coverage
• Purposive selection of respondents
– Amended method for estimating unit weights
• Sample (mis)allocation
Weighting
• Large, small weights
• Product class weights
– Category definitions
– Uncertain and dubious derivation
– Limited coverage
– Supply-Use Tables (2005 rebasing)
• Unit weights
– Purposive selection of respondents and products
– Probability proportional to size sampling?
– Consistency versus unbiasedness?
Non-Standard Estimation Methods
• Model contracts (e.g. computer programming)
• Hourly rates (e.g. legal services)
• Unit values (e.g. telecommunications)
– cheap and easy
– potential bias (but no variance)
– quality adjustment
Other Aspects
• Percentage fees (e.g. employment agencies)
– Dependence on related index
• Use of external indices
– Quality assurance
• Annual chain-linking
• Scope and coverage
– ProdCom for services? All products?
• Cost
• Timeliness
Plans to Improve and Monitor Quality
• Implement QA review recommendations
– Re-specify vague item descriptions
– Reduce processing risks
• Re-allocate sample
– Standard errors
– Re-categorisation
– 2000 weights versus 2005 weights
• Monitor unit value indices
• Periodically repeat quality assessment
• Resource constraints